Paraphrase and aggregate with large language models for improved decisions

1 . A method of interpreting a verbal input, the method comprising:

obtaining a verbal input;

assigning a meaning classification to the verbal input, and a confidence score to the meaning classification; and

based on the confidence score corresponding to the meaning classification of the verbal input being less than or equal to a threshold,

generating at least one paraphrase of the verbal input using at least one large language model (LLM);

assigning the meaning classification to the at least one paraphrase, and the confidence score to the meaning classification;

concatenating the verbal input, the at least one paraphrase, the meaning classification, and the confidence score to generate a concatenated input;

inputting the concatenated input into the at least one LLM;

selecting one of the at least one paraphrase or the verbal input, and its corresponding meaning classification, as an interpreted verbal input and an interpreted meaning classification, based on an output of the at least one LLM generated from the concatenated input; and

executing an operation based on at least one of the interpreted verbal input or the interpreted meaning classification.

2 . The method of claim 1 , further comprising:

based on the confidence score corresponding to the meaning classification of the verbal input being higher than the threshold,

inputting the verbal input into the at least one LLM;

selecting the verbal input and its corresponding meaning classification as the interpreted verbal input and the interpreted meaning classification, based on an output of the at least one LLM generated from the verbal input; and

executing the operation based on the interpreted verbal input and its corresponding meaning classification.

3 . The method of claim 1 , further comprising:

based on the confidence score of the interpreted verbal input being less than or equal to the threshold, determining that the verbal input is out-of-domain.

4 . The method of claim 1 , further comprising:

prior to concatenating the verbal input, the at least one paraphrase, the meaning classification, and the confidence score, selecting phrases from among the verbal input and the at least one paraphrase that have confidence scores exceeding the threshold,

wherein the concatenation is performed on the selected phrases and their corresponding meaning classifications and confidence scores to obtain the concatenated input.

5 . The method of claim 1 ,

wherein the verbal input is a user inquiry, and

wherein the operation executed is to generate and output a response to the user inquiry.

6 . The method of claim 5 , wherein the at least one LLM is used to generate and output the response to the user inquiry.

7 . An electronic device, comprising:

at least one processor;

at least one memory configured to store instructions that, when executed by the at least one processor, cause the electronic device to:

obtain a verbal input;

assign a meaning classification to the verbal input, and a confidence score to the meaning classification; and

based on the confidence score corresponding to the meaning classification of the verbal input being less than or equal to a threshold,

generate at least one paraphrase of the verbal input using at least one large language model (LLM);

assign the meaning classification to the at least one paraphrase, and the confidence score to the meaning classification;

concatenate the verbal input, the at least one paraphrase, the meaning classification, and the confidence score to generate a concatenated input;

input the concatenated input into the at least one LLM;

select one of the at least one paraphrase or the verbal input, and its corresponding meaning classification, as an interpreted verbal input and an interpreted meaning classification, based on an output of the at least one LLM generated from the concatenated input; and

execute an operation based on at least one of the interpreted verbal input or the interpreted meaning classification.

8 . The electronic device of claim 7 , wherein the instructions further cause the electronic device to:

based on the confidence score corresponding to the meaning classification of the verbal input being higher than the threshold,

input the verbal input into the at least one LLM;

select the verbal input and its corresponding meaning classification as the interpreted verbal input and the interpreted meaning classification, based on an output of the at least one LLM generated from the verbal input; and

execute the operation based on the interpreted verbal input and its corresponding meaning classification.

9 . The electronic device of claim 7 , wherein the instructions further cause the electronic device to:

based on the confidence score of the interpreted verbal input being less than or equal to the threshold, determine that the verbal input is out-of-domain.

10 . The electronic device of claim 7 , wherein the instructions further cause the electronic device to:

prior to concatenating the verbal input, the at least one paraphrase, the meaning classification, and the confidence score, select phrases from among the verbal input and the at least one paraphrase that have confidence scores exceeding the threshold,

wherein the concatenation is performed on the selected phrases and their corresponding meaning classifications and confidence scores to obtain the concatenated input.

11 . The electronic device of claim 7 ,

wherein the verbal input is a user inquiry, and

wherein the operation executed is to generate and output a response to the user inquiry.

12 . The electronic device of claim 11 , wherein the at least one LLM is used to generate and output the response to the user inquiry.

13 . The electronic device of claim 11 , further comprising:

a display,

wherein the operation executed is to generate and display the response to the user inquiry on the display.

14 . The electronic device of claim 11 , further comprising:

a speaker,

wherein the operation executed is to generate and play the response to the user inquiry on the speaker.

15 . A non-transitory computer readable medium comprising instructions which, when the instructions are executed by at least one processor, cause the at least one processor to carry out:

obtaining a verbal input;

assigning a meaning classification to the verbal input, and a confidence score to the meaning classification; and

based on the confidence score corresponding to the meaning classification of the verbal input being less than or equal to a threshold,

generating at least one paraphrase of the verbal input using at least one large language model (LLM);

assigning the meaning classification to the at least one paraphrase, and the confidence score to the meaning classification;

concatenating the verbal input, the at least one paraphrase, the meaning classification, and the confidence score to generate a concatenated input;

inputting the concatenated input into the at least one LLM;

selecting one of the at least one paraphrase or the verbal input, and its corresponding meaning classification, as an interpreted verbal input and an interpreted meaning classification, based on an output of the at least one LLM generated from the concatenated input; and

executing an operation based on at least one of the interpreted verbal input or the interpreted meaning classification.

16 . The non-transitory computer readable medium of claim 15 , wherein the instructions further cause the at least one processor to carry out:

based on the confidence score corresponding to the meaning classification of the verbal input being higher than the threshold,

inputting the verbal input into the at least one LLM;

selecting the verbal input and its corresponding meaning classification as the interpreted verbal input and the interpreted meaning classification, based on an output of the at least one LLM generated from the verbal input; and

executing the operation based on the interpreted verbal input and its corresponding meaning classification.

17 . The non-transitory computer readable medium of claim 15 , wherein the instructions further cause the at least one processor to carry out:

based on the confidence score of the interpreted verbal input being less than or equal to the threshold, determining that the verbal input is out-of-domain.

18 . The non-transitory computer readable medium of claim 15 , wherein the instructions further cause the at least one processor to carry out:

prior to concatenating the verbal input, the at least one paraphrase, the meaning classification, and the confidence score, selecting phrases from among the verbal input and the at least one paraphrase that have confidence scores exceeding the threshold,

wherein the concatenation is performed on the selected phrases and their corresponding meaning classifications and confidence scores to obtain the concatenated input.

19 . The non-transitory computer readable medium of claim 15 ,

wherein the verbal input is a user inquiry, and

wherein the operation executed is to generate and output a response to the user inquiry.

20 . The non-transitory computer readable medium of claim 19 , wherein the at least one LLM is used to generate and output the response to the user inquiry.